Simon Smith

Scalable solutions to meaningful problems—digital strategy, digital health and random musings

Menu Close

The future of WellPilot

It’s been awhile since I’ve written anything about WellPilot, an experimental project that uses machine learning to analyze treatments, preventions and more for hundreds of conditions. And if you were to visit, you’d notice that the site is now offline. So what happened?

Well, in summary, life happened. Our team has been overwhelmed with other projects, as well as day-to-day work and family obligations. The good news, however, is that we’re not giving up. Rather, we’re retrenching, and looking to find WellPilot a home (and funding) to realize our vision of turning complex health data into simple guidance for optimal well-being.

So keep an eye out for the future evolution of WellPilot. And if you’d like further details, or are interested in getting involved, reach out to me on Twitter.

What #DeepDream suggests about how we form ideas about beauty

I’ve been fascinated by Google’s “dreaming” AI and the new #DeepDream pictures using the algorithm and pouring into Twitter. Basically, neural nets are learning by observing lots of examples of objects, then forming idealized versions of those objects in their mind. They can then produce their vision on command, allowing others to see what they see.

It all strikes me as providing good insight into how humans form ideas about objects and concepts. Specifically, beauty. Theories of what we see as beautiful in a face suggest that signs of youth and health are key components, but possibly also “facial averageness.” Basically, our brains observe lots of people and combine the faces we see into a single average face against which we compare other people that we see.

Is this not exactly what neural nets (and, basically, most machine learning systems) are doing? It would be interesting to feed a machine learning system thousands of images of people and see if the face they perceive as average matches what most people would consider attractive. 

Image credit: Michael Tyka/Google

Fiverr, Uber and the freelance economy

I just took a trip to O’Hare airport in Chicago in an Audi A8, driven by an UberX driver who said he makes $45 an hour to help support his family, and that his son and sister now also drive for the service at his recommendation. Not only is he happy with the money, he also gets to make his own hours—and write off car expenses.

I also spent the past week testing out various services on Fiverr, to help out with my hobby project WellPilot. For example, I wanted to get a range of user feedback, so commissioned several $5 user tests. A few of them—especially a video user test—provided great value.

While the freelance economy has risks and downsides—job insecurity being a big one—my recent experiences with UberX and Fiverr appear to have been largely positive for both provider and consumer. In fact, my UberX driver today fully acknowledged the potential future of driverless cars, and how that might affect Uber’s offering (and, by extension, of course, his freelance work), but recognized that was a few years off. For now, he’s happy.

That’s not to say we shouldn’t work to mitigate the risks and downsides. But it’s hard to argue with someone making more than six times minimum wage doing something he enjoys, as his own boss.

Momentum, setbacks and hope for WellPilot

So, it’s been a long time since I posted my most recent update on WellPilot, my hobby project to turn health data into actionable, easy-to-understand guidance for well-being. Since then, we’ve had great momentum, some setbacks and some glimmers of hope:

  • Traffic on the site was growing in hockey-stick fashion for August and September, raising hopes that we would see continued growth in October. Then, suddenly, we dropped significantly month over month. I’m still unsure what happened but, given the percentage of traffic we get that’s organic, a Google algorithm change may be partly responsible.
  • On the positive side, while the site has been controversial (see next point), I have had some positive feedback and support, including from healthcare professionals, entrepreneurs and investors. I’d say the general consensus is there’s a nugget of something here that, if we can refine it and scale, has potential. But there are lots of hurdles along that path to be sure. For example…
  • We have had some legitimate criticism over our algorithmic approach to data aggregation and analysis, which I genuinely take to heart. We do want to improve our approach and make our findings as valid and reliable as possible. Some criticism, however, has gotten personal, such as suggestions that my day job—working in digital health, which includes work with pharmaceutical companies—has somehow influenced me and the algorithms in some way, such as perhaps by prioritizing pharmaceutical treatments with our algorithm. This couldn’t be further from the truth, on so many levels. For one thing, my work on WellPilot is a personal hobby that has nothing to do with my employer. For another, we’re trying to make the algorithm as unbiased as possible, and in fact include many non-pharma treatments such as supplements and even mind-body interventions. We want to find what really works, regardless of what it is or who makes it. Our algorithm isn’t perfect, but we’re continuously working to improve it—including with the help of our users, even if they’re harsh critics.
  • Fortunately, while the criticism can sting, we’ve had a tremendous outpouring of support following my recent guest post of an article describing why and how I started WellPilot, and some of the data challenges we face and are working to overcome. Several readers have gotten in touch to express support for the article and the project.

And regarding that latter bullet, I’m going to close by embedding an awesome tweet received after posting the article:

And now, back to coding.

WellPilot gets side effects smarts

WellPilot side effects index screenshotI’ve been working for a few months now on a hobby project, WellPilot, with the goal of turning complex health data into simple guidance for optimal wellbeing. (Probably need to simplify that statement, huh?)

For this hobby, I keep an eye out for useful data sources that can help make WellPilot more accurate and more useful. So I was excited when I read about openFDA, the US Food and Drug Administration’s new initiative to make their data more open and accessible.

The first dataset they’ve launched is for adverse events. An adverse event is, for the most part, a negative side effect. Healthcare professionals and manufacturers report these to the FDA for drugs, supplements and other things people tend to put in their body.

This reporting is important because many side effects aren’t discovered until a large number of patients start using a marketed product. While clinical trials are supposed to be representative of a population taking a drug, they are often too small to capture every meaningful permutation of biology (think, for example, about how genetic variations affect drug metabolism) and too short to fully understand the long-term effects. This is why post-marketing databases such as the FDA’s adverse events database are so important.

Unfortunately, such databases haven’t been easy to tap, especially for hobbyist programmers like me.

With openFDA, that’s all changed. Through a simple application programming interface (API)—basically, enabling online requests that return data in machine-usable format—the FDA has made adverse event information easily accessible.

WellPilot now incorporates this information in three ways:

Some people might wonder why this is useful. After all, many websites offer side effect information—for example, consumer health sites such as WebMD. However, these are often not comprehensive or entirely up to date. For example, the official labels for prescription drugs are based off clinical trials, with limitations noted above. WellPilot side effect data is regularly refreshed based on reports to the FDA.

This is just a start of how WellPilot might use adverse event data—not to mention other data openFDA should have in future. For example, some ideas I’m considering:

  • Allowing people to track treatments they’re taking to see of which side effects they’re most at risk overall
  • Allowing people to see both how effective a treatment is for a condition, as well as how severe the side effects might be
  • Making analysis of condition-treatment relationships smarter by informing the analysis algorithm of what’s a likely side effect (for example, with adverse event data, the algorithm would be more likely to flag “nausea” as a side effect of chemotherapy treatment rather than something chemotherapy treats)
  • Evaluating side effect risks of treatments for specific conditions (whereas we currently just evaluate it for treatments overall)

If you have any thoughts, feel free to comment here or reach out to me on Twitter.

* Note: We call them “side effects” on WellPilot rather than “adverse events” because the former is a more consumer-friendly term. In reality, however, some adverse events are not really side effects. They include things like drugs not working at all, or a method of administration not working as intended. But, for the most part, I think the label works, and it’s what consumers are most familiar with. 

Switch: My thoughts on a long-overdue “read”

SwitchWith two kids—one three, one three months—I’ve had little time to read books lately. I’m still reading loads, mind you. But mostly just articles pulled through Feedly and Zite. Finding time to sit down and read long-form narrative isn’t easy. So my books-to-read backlog has been growing.

Recently, I learned about a new audiobook service called—somewhat uncreatively— I found it while searching for a streaming audiobook service similar to Rdio and Netflix (both of which I subscribe to). My thinking: why not listen to books whenever I have the time, such as while taking my sleeping son for a walk in the stroller.

Unfortunately, didn’t quite meet my needs, as you basically get one audiobook a month for $15—and I’d like to “read” more than that. But it got me excited about the possibility of audiobooks, which led me to explore books offered through Overdrive by Toronto Public Library. I was initially skeptical, but my wife prompted me to dig deeper, and I’m glad I did. It’s a goldmine.

One of my findings: Switch, by the Heath brothers, who also wrote the excellent Made to Stick. I recently finished it and, overall, was equally as impressed as I was with Made to Stick. While some of the information wasn’t new (in particular, stuff related to ground covered in-depth by Robert Cialdini), it was woven together in a useful framework, and chock full of valuable ideas and stories that made them stick.

So, here’s my very brief summary of how to make a switch:

  1. Direct the rider (note: the “rider”—basically, the rational mind—and “elephant”—basically, emotional desires—come from the book The Happiness Hypothesis)
    1. Follow the bright spots: Often, when we’re challenged with solving a challenge, we tend to look for problems that are inhibiting the solution. But there’s another approach: find what is working and try to replicate it. For example, if you want customers to buy more products, you might focus on all the myriad reasons they’re not buying more from you. Another approach would be to identify customers who are buying more from you, and see why it’s working. Then try to replicate it.
    2. Script the critical moves: “Eat healthier” is a vague statement with a lot of wiggle room. (Especially since what’s considered healthy differs depending on which fad diet might be popular this month.) “Drink two glasses of water every day as soon as you wake up” is a specific action. Often, people get stuck in analysis paralysis, or get overwhelmed with options. To overcome this, script the critical moves: tell them exactly what to do, and make it specific, with black-and-white edicts that leave no wiggle room.
    3. Point to the destination: What does success look like? Paint a clear picture of what will happen if people act out the critical moves. For example, continuing with the healthier eating theme, you could have them purchase an outfit that’s the size they want to fit into, so they can clearly visualize what success will look like.
  2. Motivate the elephant
    1. Find the feeling: If, like me, you work in marketing, you understand this implicitly. You can’t rationally convince people to change their behavior, or do what you want them to do. You need to make them feel something. Once you hook them with feeling, you can talk to them intellectually. Exactly what the feeling is will depend on the change you’re trying to accomplish. If we stick with the health theme, telling people “eat healthier or your cholesterol will go up” won’t be as effective as talking to people about how sad it would be for them to die of a heart attack and miss their children’s graduations and weddings, and the birth of their grandchildren. A better line for a campaign might be: “Cut out red meat. Your grandchildren want to meet you.”
    2. Shrink the change: One of the great examples in this book—and there are many—talks about a study of customer loyalty cards that found a card for a car wash requiring 8 stamps for a free wash performed worse than a card requiring 10 stamps but with 2 pre-stamped. (That’s a mouthful. Sorry. No time for much editing these days either!) The reason? Big changes spook the elephant. While you still needed 8 washes to get a free wash, in the former case you needed to fill 100% of your stamps, while in the latter you were already 20% of the way there. Going back to our health examples, we might ask people to eat just one more fruit per day. It seems like a small thing, but it’s the kind of change that’s easy to do, and ultimately small changes can snowball into big ones.
    3. Grow your people: This topic is one that dovetails with Robert Cialdini’s discussions of commitment and consistency. Just as you can make change seem easier by shrinking it, you can also make it seem easier by growing the people who will tackle it. A lot of this comes down to identity. For one thing, people who think of themselves as capable of making change are more likely to make it. So, for example, your company can rebrand everyone in the organization “change engineers” or “innovators,” rather than the dull “employees.” Another key point here is, as Cialdini noted, that people will act in ways that are consistent with their identity. So if you find the bright spots (see above), and point out to people that they have previously done things in alignment with the changes you want to make, they will subsequently be more likely to make similar changes. For example, again going back to the topic of health, if you give people a FitBit that shows they’re already hitting 7,500 of their target 10,000 steps a day without every stepping foot in a gym, they’ll start to see themselves as physically active and be more likely to walk the extra steps to hit the 10,000. (This example also illustrates that the information helps shrink the change, because they were already 75% of the way to their goal and didn’t even know it.)
  3. Shape the path
    1. Tweak the environment: The Heath brothers repeat, and rightfully so, that what often seems like a people problem is actually a situation problem. So if you can change the situation—and especially the environment—you can often change people’s behavior. A simple example for health is simply eliminating all junk food from someone’s fridge and cupboards to make it more difficult for them to eat unhealthy snacks. Less willpower is needed when temptation has been removed.
    2. Build habits: It can take time to create habits, which means changed behavior can often slip back. However, there are ways to instil instant habits using “action triggers.” For example, in the book, the Heath brothers talk about a safety supervisor at a manufacturing plant who painted a blue line around the perimeter, within which safety gear had to be worn at all times. The blue line acted as an action trigger, prompting people to put on their safety gear habitually. In relation to health, an action trigger could be setting out your gym clothes the night before, prompting you to put them on when you wake up and head to the gym.
    3. Rally the herd: This topic also overlaps with Cialdini, who talks about the importance of social proof in influencing people’s behavior. Basically, we look to others for behavioral guidance, and do this more often when things are uncertain. Since change creates lots of uncertainty, people will look to others and see how they’re reacting. If nobody else is making the change, you can be pretty sure it won’t catch on. For this reason, you can’t just think about individuals when trying to create change. You need to think about social cues that people will give to each other, and ensure alignment. For example, if we wanted to influence 40-year-old men to see their doctor for a cholesterol test, we could promote something like, “80% of men 40 and over see their doctor for a cholesterol test—have you?” Perhaps we could also use a social media campaign with checkins at the doctor for a cholesterol test, demonstrating that this is something people’s friends are doing.

All in all, Switch was a worthwhile read—um, I mean, listen—and I’m thrilled that audiobooks can allow me to start consuming great books again. I’m now working my way through Superfreakonomics, so I may have another blog post book review coming soon.

Fancy Hands: after two months, what’s my take on this virtual personal assistant service?

Disclaimer: Links to Fancy Hands within this article are special referral links—effectively a Fancy Hands coupon. If you click the links and sign up, you get 50% off your first month and I get a $10 credit. This doesn’t really help me unless I continue to think the service is great and continue using it, so it in no way affects my opinions below. But I’m noting it for full transparency. Now, on to the post. 

Two months ago, I wrote about a new service I’m using called Fancy Hands. The service makes virtual assistants app-like. You literally get an app on your mobile device to which you post requests that virtual personal assistants fulfill. At the time, I had just started using it. So what’s my experience after two months?

Since I’ve started using Fancy Hands, I’ve used it to complete 34 requests. For these, they’ve placed 32 calls, sent 4 emails, saved me $10 (by finding discounts for things I want them to buy), spent 54 minutes on the phone and saved me 16.8 hours (by their calculations). I believe my total cost for their services was $115, or about $6.85 per hour saved. (And yes, the fact that they report all of this information on my dashboard makes a quantified-selfer like me quite happy.)

So, quantitatively, Fancy Hands continues to make sense to me. In fact, along the way I’ve increased my service level from 5 tasks per month to 15. And for now, that seems about the right number. So what am I primarily using it for?

  • Gift purchases. I think Fancy Hands has made me more thoughtful, because no longer is cash simply my default gift due to time constraints. I can tell Fancy Hands about the person I want to buy gifts for, they can recommend good options for these people, and then they can purchase the ones that I approve.
  • General purchases. Sometimes, I need a specific item but don’t have time to go searching for it. I can ask Fancy Hands to find me options for those items, including those that are highly rated, and then approve them for purchase. One advantage here is that they can also search for or come across discounts, thereby helping to save me money overall.
  • Research and data entry. Overall, I’ve found the quality of assistants to be very high. On several occasions I’ve asked them to conduct research, including in fairly specialized subjects, and they’ve returned good results. They’ll also put those results into structured data formats such as spreadsheets. This can be a huge time-saver. My sense is that Fancy Hands may use a person or algorithm to get the right requests to the right people. For example, if an assistant has expertise in digital media, they might direct a related request to them. (But this is just an assumption.)
  • Finding and booking contractors. One thing I wish Fancy Hands offered was physical as well as virtual service. For example, if I want to change all the lightbulbs in my house to LEDs, it would be great if I could post that request somewhere and have it done. (I’ve tried to find a good service for this in Toronto, but those that exist do not seem to have much traction.) However, Fancy Hands can help me find, book and post ratings for contractors, which I have had them do with good success.
  • Feedback. Sometimes, you need feedback. For example, I’m currently working on a hobby project called WellPilot (I wrote more about this here and here). I’ve asked Fancy Hands to give me feedback on aspects of the site such as usability. This feedback has been very useful.
  • General digital tasks. One that was particularly useful for me was having Fancy Hands add my digital signature to a huge batch of documents that I needed to sign related to a business transaction. I had reviewed the documents in PDF format, but would have had to print, sign and fax each one. Instead, I sent them to Fancy Hands with a digital signature and had them add the signatures for me.

Overall, I would (and regularly do) recommend Fancy Hands to anyone who’s trying to optimize their time. Fancy Hands’ motto is “do what you love and let us do the rest.” I plan to continue exploring how to best make that happen, and keep looking for opportunities to outsource to them.


Making WellPilot smarter: scoring condition-treatment relationships

A few weeks ago I posted about WellPilot, a hobby project to evaluate what really works for maintaining health and extending healthy lifespan. At that time, it was pretty basic in its evaluation of condition-treatment relationships. It simply counted the quantity of research articles supporting a relationship, but ignored their quality. For example, a cell culture experiment would be counted as equal to a randomized, double blind, placebo-controlled trial.

I’ve now addressed this with a more sophisticated condition-treatment scoring algorithm. For example, if you look at what WellPilot has discovered about Parkinson’s disease and melatonin, you’ll see that the highest scored article is randomized, double blind and placebo-controlled. WellPilot has now scored over 1.5 million research articles across nearly 9,000 condition-treatment relationships. (I set up a WellPilot Twitter account where it now randomly announces different relationships that it’s analyzed.)

However, there’s still much to do to make WellPilot’s results more meaningful and reliable. For one thing, it doesn’t know whether an article reflects a positive or negative finding for a condition-treatment relationship. It just knows the quantity and, to an extent, the quality of the studies conducted. One of my next areas of focus is helping to train the system by involving users in crowdsourcing feedback on research articles supporting a condition-treatment relationship. WellPilot could then, theoretically, be programmed to learn what signals to look for in evaluating research. My hobby project to guide health maintenance (and possibly life extension)

I’m a bit of a health nut. I pop lots of supplements. I’m also into life extension, and am signed up for cryonics.

So it was a bit frustrating recently when, after my wife and I completed a 23andMe genetic test, we found no easy-to-use sources of information on how to prevent some of the conditions for which we were genetically at risk.

I think that 23andMe is working to close this gap with the acquisition of CureTogether. But that site uses patient self-reporting. I like the idea of crowdsourcing treatments, but also like the idea of using more traditional data sources. And I wish someone could assemble an end-to-end treatment and prevention platform, including data-driven guidance, subscription-based products (to avoid the annoyance of reordering), and integrated testing to ensure the interventions are working.

Nerd that I am, I’ve started to take a stab at this. It’s a hobby project that I call WellPilot. The initial functionality works like this:

  • Search WellPilot for conditions or treatments
  • Click on a condition  (for example, Parkinson’s), and see what treatment and prevention is supported by research (currently just non-drug interventions)
  • Or click on a treatment (for example, Green tea), and see what conditions it can treat or prevent, according to research

You can also then click through and see all the research supporting condition-treatment relationships (for example, coffee for Parkinson’s disease).

Currently, the data is imported from PubMed, and the analysis is very rudimentary—it uses research volume for each condition-treatment pair. Moving forward, I aim to improve this, to account for such factors as study type (human, animal, cell, etc.), study strength (randomized, placebo controlled, etc.) and reputation (impact factor, etc.).

I would also like to eventually add crowdsourcing, similar to CureTogether, but that’s a longer-term ambition. And I’d like to add product subscriptions—to provide premium supplements at wholesale prices, delivered to your door, based on strong evidence. And the list of feature ideas keeps growing.

For now, though, it at least provides a starting point if you want some guidance for what interventions can address conditions you’re concerned about. Check it out and let me know what you think.

Fancy Hands review: finally, a virtual assistant that’s as easy to use as an app (literally)

The Fancy Hands app is like a magic to-do list: add stuff to it, and other people get it done

The Fancy Hands app is like a magic to-do list: add stuff to it, and other people get it done

Disclaimer: Links to Fancy Hands within this article are special referral links—effectively a Fancy Hands coupon. If you click the links and sign up, you get 50% off your first month and I get a $10 credit. This doesn’t really help me unless I continue to think the service is great and continue using it, so it in no way affects my opinions below. But I’m noting it for full transparency. Now, on to the review. 

A few weeks ago, in one of my regular rants about personal effectiveness, I half-jokingly told my wife that I wanted a to-do list that worked like magic: I would add stuff to it, then other people would get that stuff done for me.

The universe granted my wish.

A few days later, reading an article on several businesspeople’s killer apps for personal effectiveness, I read about Fancy Hands. It sounded almost exactly like what I had described: post a request, and a virtual assistant gets it done (with few limitations—most critically, the task must be completable with either a computer or a phone, as they don’t do physical tasks yet).

I was intrigued. I have often thought about hiring a personal assistant, inspired by books like 4-Hour Workweek to outsource anything that doesn’t require my direct effort to complete. But I’ve yet to find a service that’s cost-effective and efficient. Most virtual assistant services require an upfront purchase of hours, which can be a significant cost—especially if you fail to use them all. And if I want to hire a personal assistant directly, it can get expensive—plus, being just one person, they can’t scale to tackle multiple tasks simultaneously when needed. For one-off tasks, I’ve used Fiverr, but it’s not efficient for regular use (it takes too much time to find the right people you can trust), and their mobile experience leaves much to be desired.

So Fancy Hands seemed just right. They offer subscription packages, with 5 tasks for $25 being the starting point. While this works out to $5 per task, it drops to about $2.60 per task at the 25 tasks for $65 level. Plus, from my experience, some tasks—such as scheduling appointments—are free. And the ease of use is superb: through either the app or the website, just submit a request and someone acts on it.

Buying time

So I signed up. And, while my wife often thinks it’s nuts, I’ve since used Fancy Hands to research a new shampoo for my daughter, reschedule a hair appointment, buy me shirts (they’ll purchase anything up to $100 on your behalf), do research for a business venture, do research for floor treatment, and research and purchase a tongue pad to help a new pair of shoes fit better.

All in all, their dashboard tells me I saved 30:13 with those requests. At $25, that works out to about $0.83 a minute—which, if I want to compare, is significantly less than the rate I can charge for consulting work, meaning there is plenty of ROI potential if I use Fancy Hands to liberate time for paid work.

And the quality? Overall, excellent. For almost every request, Fancy Hands’ assistants go above and beyond, following my directions exactly or recommending alternatives if my original request can’t be fulfilled for reasons beyond their control (such as a product I asked for not existing). While I’m still working to identify the best tasks to delegate, I have few qualms that requests I submit will be appropriately addressed.

I’ve also noticed some more intangible benefits. For one, I feel more at ease just knowing Fancy Hands is available—it’s like a pressure valve when work and family obligations build up. Another is that it’s training me to better delegate, as I’m a perfectionist and tend to avoid delegating due to fears about quality and loss of control. And it’s forcing me to regularly think about the value of time, and how my time can be best spent (note: the founder of Fancy Hands, Ted Roden, tries to never do the same thing twice).

Bottom line: subscribing to Fancy Hands can buy you time. While there’s certainly room for improvement—I wish they could run physical errands, for example—I’m definitely planning to continue my subscription, and probably increase it. So far, they’re living up to their motto: do what I love and let them do the rest.

© 2015 Simon Smith. All rights reserved.

Theme by Anders Norén.

© 2000-2015 Simon Smith All Rights Reserved